Ship and explain the bounded research agent

A shippable lab artifact includes: how to run it, the metric result (step count ≤ max; citation coverage; abstention on empty search), the failure you can still reproduce (infinite tool loop, or final answer with no citations after search), the security gate for agent inventing tools outside the registry mid-run, and a rollback note. The user decision it supports remains: search a local knowledge base then produce a cited answer within a step budget.

RUN_LIVE_EVALS=1 npm run test:live

Expected evidence: **The agent searches once, cites kb://returns, and stops before MAX_AGENT_STEPS.**. Store this beside the fixture version so scores remain meaningful after content changes in `agents-in-code`.

State operating limits for the bounded research agent in plain language: fixture size, offline vs live dependencies, and what would require a new eval set. Shipping agents-in-code is honest scoping, not maximal confidence language.

Fill this table in your notes for the bounded research agent:

• Proved on fixture KB with 4 notes + max_steps=3: …
• Unproved beyond the fixture: …
• Metric that blocks release: step count ≤ max; citation coverage; abstention on empty search
• Failure still reproducible: infinite tool loop, or final answer with no citations after search
• Security gate: agent inventing tools outside the registry mid-run
• Rollback: …

Ship the narrative only when the unproved list is honest. Reviewers trust narrow claims that support search a local knowledge base then produce a cited answer within a step budget more than maximal language that collapses under the first production oddity.

Hand your ship note to a peer and ask them to recreate a proved/unproved ship note with rollback without watching you type. If they cannot, your evidence is still tribal knowledge. Tighten the run command and the metric line until a stranger can validate the bounded research agent against fixture KB with 4 notes + max_steps=3.

At the mastery and shipping stage for agents-in-code, the job is narrower than finishing a product demo. You are creating one progressive evidence piece about fixture KB with 4 notes + max_steps=3 that later pages inherit without redefining success. Keep that fixture small enough to inspect by hand, keep outputs copy-pasteable as text, and refuse to narrate this baseline as if it were a production SLA: single-shot LLM answer without search.

For this page specifically, success looks like a proved/unproved ship note with rollback while still centering the user decision to search a local knowledge base then produce a cited answer within a step budget. If you cannot point to a file, command, or assertion that proves that for the bounded research agent, stay on this page instead of advancing.

Glossary: tool · Glossary: structured output · Cheatsheet: production ops signals

Previous